我想同时使用 Boost.Asio 的链和优先包装器。
在编写代码之前，我已经阅读了以下信息：
http://thread.gmane.org/gmane.comp.lib.boost.asio.user/3531
为什么在使用 boost::asio 时每个连接都需要 strand？
我想使用包装器方法，因为我想使用各种异步 API，例如 async_read、async_write 和 async_connect。根据http://thread.gmane.org/gmane.comp.lib.boost.asio.user/3531，似乎可以组合优先级包装器和链包装器。
所以我根据下面的例子写了代码：
这是我的代码：
#include <iostream>
#include <functional>
#include <queue>
#include <vector>
#include <thread>
#include <mutex>
#include <boost/asio.hpp>
#include <boost/optional.hpp>
#define ENABLE_STRAND 1
#define ENABLE_PRIORITY 1
class handler_priority_queue {
public:
template <typename Handler>
void add(int priority, Handler&& handler) {
std::cout << "add(" << priority << ")" << std::endl;
std::lock_guard<std::mutex> g(mtx_);
handlers_.emplace(priority, std::forward<Handler>(handler));
}
void execute_all() {
auto top = [&]() -> boost::optional<queued_handler> {
std::lock_guard<std::mutex> g(mtx_);
if (handlers_.empty()) return boost::none;
boost::optional<queued_handler> opt = handlers_.top();
handlers_.pop();
return opt;
};
while (auto h_opt = top()) {
h_opt.get().execute();
}
}
template <typename Handler>
class wrapped_handler {
public:
wrapped_handler(handler_priority_queue& q, int p, Handler h)
: queue_(q), priority_(p), handler_(std::move(h))
{
}
template <typename... Args>
void operator()(Args&&... args) {
std::cout << "operator() " << std::endl;
handler_(std::forward<Args>(args)...);
}
//private:
handler_priority_queue& queue_;
int priority_;
Handler handler_;
};
template <typename Handler>
wrapped_handler<Handler> wrap(int priority, Handler&& handler) {
return wrapped_handler<Handler>(*this, priority, std::forward<Handler>(handler));
}
private:
class queued_handler {
public:
template <typename Handler>
queued_handler(int p, Handler&& handler)
: priority_(p), function_(std::forward<Handler>(handler))
{
std::cout << "queued_handler()" << std::endl;
}
void execute() {
std::cout << "execute(" << priority_ << ")" << std::endl;
function_();
}
friend bool operator<(
queued_handler const& lhs,
queued_handler const & rhs) {
return lhs.priority_ < rhs.priority_;
}
private:
int priority_;
std::function<void()> function_;
};
std::priority_queue<queued_handler> handlers_;
std::mutex mtx_;
};
// Custom invocation hook for wrapped handlers.
template <typename Function, typename Handler>
void asio_handler_invoke(Function&& f,
handler_priority_queue::wrapped_handler<Handler>* h) {
std::cout << "asio_handler_invoke " << std::endl;
h->queue_.add(h->priority_, std::forward<Function>(f));
}
//----------------------------------------------------------------------
int main() {
int const num_of_threads = 4;
int const num_of_tasks = 5;
boost::asio::io_service ios;
boost::asio::strand strand(ios);
handler_priority_queue pq;
for (int i = 0; i != num_of_tasks; ++i) {
ios.post(
#if ENABLE_STRAND
strand.wrap(
#endif
#if ENABLE_PRIORITY
pq.wrap(
i,
#endif
[=] {
std::cout << "[called] " << i << "," << std::this_thread::get_id() << std::endl;
}
#if ENABLE_PRIORITY
)
#endif
#if ENABLE_STRAND
)
#endif
);
}
std::vector<std::thread> pool;
for (int i = 0; i != num_of_threads; ++i) {
pool.emplace_back([&]{
std::cout << "before run_one()" << std::endl;
while (ios.run_one()) {
std::cout << "before poll_one()" << std::endl;
while (ios.poll_one())
;
std::cout << "before execute_all()" << std::endl;
pq.execute_all();
}
}
);
}
for (auto& t : pool) t.join();
}
包装器由以下宏启用：
#define ENABLE_STRAND 1
#define ENABLE_PRIORITY 1
启用两个宏后，我得到以下结果：
before run_one()
asio_handler_invoke
add(0)
queued_handler()
before poll_one()
asio_handler_invoke
add(1)
queued_handler()
asio_handler_invoke
add(2)
queued_handler()
asio_handler_invoke
add(3)
queued_handler()
asio_handler_invoke
add(4)
queued_handler()
before execute_all()
execute(4)
execute(3)
execute(2)
execute(1)
execute(0)
before run_one()
before run_one()
before run_one()
我希望我得到
[called] priority,thread_id
输出为
[called] 1,140512649541376
但我没明白。
似乎在函数execute()中，function_()被调用但wrapped_handler::operator()未被调用。（在我的代码execute()中调用了该函数。）pq.execute_all();
void execute() {
std::cout << "execute(" << priority_ << ")" << std::endl;
function_(); // It is called.
}
template <typename Handler>
class wrapped_handler {
public:
template <typename... Args>
void operator()(Args&&... args) { // It is NOT called
std::cout << "operator() " << std::endl;
handler_(std::forward<Args>(args)...);
}
function_()我在调用之后跟踪了序列。
调用以下函数：
https://github.com/boostorg/asio/blob/boost-1.63.0/include/boost/asio/detail/wrapped_handler.hpp#L191 https://github.com/boostorg/asio/blob/boost-1.63 .0/include/boost/asio/detail/wrapped_handler.hpp#L76 https://github.com/boostorg/asio/blob/boost-1.63.0/include/boost/asio/strand.hpp#L158 https:// /github.com/boostorg/asio/blob/boost-1.63.0/include/boost/asio/detail/impl/strand_service.hpp#L55 https://github.com/boostorg/asio/blob/boost-1.63。 0/include/boost/asio/detail/impl/strand_service.ipp#L94
然后在函数bool strand_service::do_dispatch(implementation_type& impl, operation* op)中，操作op没有被调用，而是被推入队列中，如下一行：
我不确定为什么将function_()其发送到 strand_service。我认为在我的代码中的以下点已经解开了绞线包装器：
template <typename Function, typename Handler>
void asio_handler_invoke(Function&& f,
handler_priority_queue::wrapped_handler<Handler>* h) {
std::cout << "asio_handler_invoke " << std::endl;
h->queue_.add(h->priority_, std::forward<Function>(f));
}
如果我只启用优先级包装器，我会得到以下结果。似乎按我的预期工作。
before run_one()
asio_handler_invoke
add(0)
queued_handler()
before poll_one()
asio_handler_invoke
add(1)
queued_handler()
asio_handler_invoke
add(2)
queued_handler()
asio_handler_invoke
add(3)
queued_handler()
asio_handler_invoke
add(4)
queued_handler()
before execute_all()
execute(4)
operator()
[called] 4,140512649541376
execute(3)
operator()
[called] 3,140512649541376
execute(2)
operator()
[called] 2,140512649541376
execute(1)
operator()
[called] 1,140512649541376
execute(0)
operator()
[called] 0,140512649541376
before run_one()
before run_one()
before run_one()
如果我只启用了 strand wrapper，我会得到以下结果。似乎也像我预期的那样工作。
before run_one()
[called] 0,140127385941760
before poll_one()
[called] 1,140127385941760
[called] 2,140127385941760
[called] 3,140127385941760
[called] 4,140127385941760
before execute_all()
before run_one()
before run_one()
before run_one()
有任何想法吗？